Gradient Boosting Machine based Fault Diagnosis in Power Electronics Circuits

In recent years, power electronics has seen significant advancements, driven by growing demand for efficient energy solutions in renewable energy, electric vehicles and smart grids. Existing approaches for diagnosing faults had faced challenges like difficulty in detecting subtle changes in system behaviour. Therefore, this research proposes Gradient Boosting Machine (GBM) for diagnosing faults in power electronic circuits. Initially, the data is collected from fault history data of single-phase Pulse Width Modulation (PWM) rectifier. which incorporates 160 samples of normal state and 40 samples of fault state. After that, Synthetic Minority Oversampling Technique (SMOTE) is employed for balancing the raw data which is collected from fault history. The relevant features of balanced data are selected by using the ReliefF algorithm. These selected features are given as input to GBM for diagnosing faults in single-phase PWM rectifier. The proposed GBM effectively achieved better results in terms of accuracy (0.9840) and F1 Score (0.9769) respectively when compared with existing Random Vector Functional Link Network (RVFL).